Time series data of wind pressure from wind tunnel test on rectangular buildings were used for the population. An appropriate probabilistic distribution on this population was estimated by applying 4 types, i.e. Normal distribution, Gamma distribution, Gumbel distribution and Weibull distribution. The method which is to estimate peak external pressure coefficient by using such characteristic values of probabilistic distribution as mean and root mean square, is suggested. The peak external pressure coefficient estimated by probabilistic distribution is compared to the value from the conventional method, which extracts the largest among moving averaged time series data of wind pressure, and the differences between those values are discussed. In case that time series data of population are divided into arbitrary estimating time, converted into subset data, and the number of estimating samples is decreased, it is examined how peak external pressure coefficient and non-exceedence probability which is required to estimate peak external pressure coefficient, are changed in relation to estimating time or number of subset.